It is increasingly common in empirical research to merge data sets containing different units of observation. When the units are not nested, a crosswalk specifying how the units from one data source are allocated to the units of the other is needed. Unfortunately, most crosswalks are ad hoc, a fact that is often ignored by researchers and has not caught the attention of econometricians. Here, I show that use of an incorrect crosswalk induces measurement error that is necessarily nonclassical and can be consequential. I discuss and illustrate the ramifications of using a flawed crosswalk, present two specification tests, offer potential solutions, and provide an application to the effects of social media on political polarization.